Messages in this thread | | | Date | Tue, 10 Sep 2013 09:23:51 -0700 | From | John Stultz <> | Subject | Re: kernel deadlock |
| |
On 09/10/2013 12:29 AM, Ingo Molnar wrote: > * John Stultz <john.stultz@linaro.org> wrote: > >> Now, I'm still in the dark as to why HRTICK exposes this, but seems like >> the following patch should resolve the issue (and quiets the lockdep >> warnings in my testing). Let me know how it works for you! >> >> Ingo: This makes me think we really should have some lockdep smarts >> added to seqlock/seqcount structures. Is there something already >> discovered thats preventing this, or has this just not yet been tried? > There's no fundamental reason to not do it, it's a very good idea! > >> From 09885df9734aa678ccb61dbef356fea4faeff16c Mon Sep 17 00:00:00 2001 >> From: John Stultz <john.stultz@linaro.org> >> Date: Mon, 9 Sep 2013 13:07:18 -0700 >> Subject: [PATCH] timekeepeing: Fix HRTICK related deadlock from ntp lock changes > Cool fix! I think I've seen this deadlock in various incarnations but was > never able to track it down. Do you have a guesstimate about roughly when > it was introduced? Is it an ancient bug?
3.10, I believe. So no not ancient, but we've seen and resolved similar issues in the past (the old printk while holding xtime_lock deadlock).
Hopefully the lockdep enablement helps sort these out for good.
thanks -john
| |